Hermes Agent Hosting
Your personal AI agent, online 24/7. We keep Hermes running, remembering, updated and isolated. You decide how it thinks: bring your own AI account, connect Nous Portal, or buy AI usage from us.
- Always on, never sleeps
- Memory that persists
- Telegram, Discord, Slack
- Daily backups you can download
Plans How it thinks What you get Isolation Should you self-host? FAQ
Hosting plans
This price is for keeping the agent alive: the runtime, the memory, the storage, the backups and the isolation. AI usage is separate and is not marked up into these numbers, because an autonomous agent's inference bill has no ceiling and hiding it inside a flat fee would mean either overcharging quiet users or absorbing losses on busy ones. What we tier is what actually costs us something.
Personal
1 agent · 1 task at a time · 10 GB storage
One agent that is always there. Backups kept 3 days.
Power Most popular
1 agent · 3 concurrent tasks · 25 GB storage
For an agent that actually works: parallel tasks, heavier automations, 14 days of backups.
Fleet
3 agents · 6 concurrent tasks · 75 GB storage
Several agents at once, for people running a small fleet. 30 days of backups.
Scheduled jobs, Telegram, Discord, Slack, skills and unlimited conversations are on every plan. Gating those would be a pricing trick, not a resource limit: they cost us nothing extra and withholding them would only make the cheap plan worse on purpose.
How your agent thinks
Hosting and inference are two different purchases. Pick whichever suits you: the hosting price does not change either way.
Connect Nous Portal
One click and it works. Nous bills the models and gives the agent web search, a cloud browser, image generation and speech. Nothing for you to configure and no API key to paste.
Use your own AI account
OpenAI, Claude, OpenRouter or anything OpenAI-compatible. Your provider, your bill, your rate limits. If the agent has an expensive afternoon, it happens on an account you control.
Supercraft AI
No external account. Top up a balance and pick a model. We bill what the model actually costs plus a small routing fee, and the agent stops when the balance runs out. No token maths, no surprise invoice.
Your own endpoint
Point it at your Ollama box, a company inference endpoint or a GPU server elsewhere. We host the agent, you host the model.
We do not sell AI by the message or by the token. The same amount of agent work can cost a couple of dollars on a small model and forty on a large one, so a "1,000 messages" allowance would mean something different every month. A balance in dollars is the only unit that stays honest.
What you actually get
| Always on | The agent runs continuously. It does not spin down between messages, so scheduled jobs fire and it is there when you message it at 3am. |
|---|---|
| Memory that persists | What it learns, the skills it writes, its personality and its project context all live on disk and survive restarts and updates. |
| Messaging | Telegram, Discord, Slack and email. Connect once and talk to it from your phone. |
| Scheduled work | Cron jobs and subagents, so it can do things while you are asleep rather than only answering when asked. |
| Backups you can take with you | Daily snapshots of the whole agent, downloadable. If you want to move it elsewhere, take it. An agent you cannot export is not really yours. |
| Updates | We track upstream releases and update the runtime. Your memory, skills and settings carry across. |
| Bring an existing agent | Already running OpenClaw? Import it: memory, skills, personality and settings come over. |
Why isolation matters here
An agent worth running can execute shell commands, run Python, browse the web and install things. That is the point of it, and it is also why this cannot be run the way a game server is. Every agent gets its own sandboxed container with its own storage, no access to the host filesystem, and capped CPU, memory and processes. Your agent cannot see anyone else's, and nobody else's can see yours.
Provider keys get the same treatment. If you bring your own OpenAI or Claude account, the key is held in our vault and the agent talks to a proxy that attaches it upstream. The agent never holds the credential, so a web page that talks it into misbehaving has nothing to steal.
Should you just self-host it?
If you are comfortable on a Linux box, then genuinely yes. Hermes is open source and runs on a cheap VPS, and if you already keep one running you do not need us. We would rather say that than sell you a wrapper around something you can do yourself.
This is for the other case: you want the agent, not the sysadmin work. Somebody has to keep it patched, keep the disk from filling, notice when it dies at 4am, take backups that actually restore, and keep an agent with shell access away from everything else on the machine. That is the job we are charging for.
Questions
Is the AI included in the price?
No, and that is deliberate. The plan price covers hosting only. You either bring your own AI account, connect Nous Portal, or top up a Supercraft AI balance. Bundling inference into a flat fee would mean charging quiet users for busy ones.
What happens if my agent goes haywire and spends a fortune?
On your own provider account, your existing limits apply. On a Supercraft balance, it stops when the balance is gone; there is no overdraft and no invoice afterwards. This is exactly why we keep the two purchases apart.
Can I use a different agent than Hermes?
OpenClaw as well, and we expect to add others. The hosting is the product; the agent is a template running on it. If something better appears next year you should be able to switch without leaving.
Do I need to know how to code?
No. Pick a plan, choose how it thinks, connect Telegram or Discord, and talk to it. The technical options are there if you want them and ignorable if you do not.
Can I take my agent elsewhere?
Yes. Backups are downloadable and include memory, skills, personality and configuration. We would rather you stay because leaving is easy than because it is hard.
When can I get one?
Not yet. This page exists to find out whether enough people want it, and how they want the AI handled, before we build it. Reserve a price and tell us how you would power it. Nothing is charged now, and nothing is charged automatically later.